Pseudo Random Numbers

[NoCforMe]

I guess I was thinking of the die-hard DOS programmer pecking away on their still-running PC or XT, green-screen monitor and everything.

[sinsi]

It's perfectly valid to use a 32-bit register to access memory in real mode, you just have to be careful that the high 16-bits are zero,
e.g. MOV ECX,[EAX+EDX*4+1000h] is OK (assuming the address computes to <64K).
Like Steve said, instead of mucking around with DX:AX for 32-bit numbers we can use one 32-bit register or (gasp) 64-bit maths with EDX:EAX 

The assembler has to support 32-bit instructions, too.

[tda0626]

This has me perplexed. Wrote a little program to output some random numbers but it is doing some weird stuff right after my

Code Select Expand

div dx instruction. It starts randomly executing code right after the instruction, see pictures below. The unassembled code segment shows my

Code Select Expand

add dx, 30 but never executes it. I have attached my source to this post too. What in the world is happening?

[sinsi]

You are trying to divide DX:AX by DX, in this case 000a0012h / 000ah, which overflows.
Use a register other than DX.

[NoCforMe]

Wait a second:
I understand the overflow problem.
But isn't it perfectly valid to divide by DX, provided it doesn't result in an overflow?
In other words, the problem isn't dividing by DX, it's the values being divided, right?
Scratch that. (E)DX is the last register you'd want to use as a divisor. Because it holds the high word/dword of the dividend.

You're right. Don't use it!

[tda0626]

Does the same thing when I change it to CX. I was under the assumption that it would return AX=0 if the number was less than 10 and put the remainder in DX. Guess not.

[NoCforMe]

Did you remember to set DX to zero? Remember, DIV does DX:AX / {dividend}.

[tda0626]

Did you remember to set DX to zero? Remember, DIV does DX:AX / {dividend}.

LOL these little mistakes drive you crazy but I am learning from them. You are right. I XOR DX, DX and it works somewhat but doesn't print out the right characters. Prints out a triangle and some other garbage. The correct values are written to the memory location though when I dump the memory. DOS function AH=9 21h just doesn't print the right stuff.

Code Select Expand

numstring DB 0,0,32,'$'
               
                mov ah, 9
mov dx, offset numstring
int 21h

[NoCforMe]

... and of course with INT 21/AH=9 you need to remember to terminate the string with a "$".

(What a weird terminator; how'd they come up with that? What if you want to print a dollar sign in the string?)

[tda0626]

I am on a roll tonight with blunders  . I was adding 30 decimal to make it an ASCII number instead of 30h, which was the reason for the weird characters. Now just need figure out why my loop exit check isn't working.

[NoCforMe]

Easy way to remember the binary--> ASCII numeric conversion thing:

Code Select Expand

  MOV  AL, number
  ADD  AL, '0'
No need to remember any hexadecimal #s.

[sinsi]

Now just need figure out why my loop exit check isn't working.

You start with CX=1000 and increment it after each loop, did you mean to decrement it?
I'm not sure if the sign flag is the one you should be testing, maybe the carry or zero flag?

[tda0626]

Yep I saw that earlier and fixed it. Got it to work finally but the period is only 15 before it starts repeating again so it does not work. Guess I need to maybe do another method or read up more on LFSR.

[NoCforMe]

About your loops:
You might want to use the LOOP instruction.
When you know exactly how many iterations you want, it makes things much easier.
You set (E)CX to the loop count.
At the bottom of the loop you code the LOOP instruction with the jump target of the loop.
LOOP automagically decrements (E)CX, compares it to zero, jumps to the loop target if not.
Saves you a little bit of bookkeeping.

Code Select Expand

    MOV   CX, loopCount
lp: ; do stuff
    ; do more stuff
    LOOP  lp
Only "gotcha" is that it's only good for short distance loops (max. 127 bytes of code).

There are even 2 more flavors of this instruction, LOOPE/Z and LOOPNE/Z that will loop while equal (zero flag set) or not equal, from an operation that sets that flag before the LOOP instruction.

The 8086/8088 was an awesome little machine.

[zedd151]

You might want to use the LOOP instruction.
...You set (E)CX to the loop count.
Only "gotcha" is that it's only good for short distance loops (max. 127 bytes of code).

And you cannot use CX within the loop (without some form of preserving it of course)...